Convolutional neural networks for detection, segmentation and classification
Convolutional neural networks are the new tool to solve almost any image analysis problem. Their flexible network layout and highly customizable loss functions allow extremely fast detection, segmentation and tracking in 2D and 3D images.
Invariant features for detection, segmentation and classification
Invariant features allow robust detection and classification independent of specimen orientation and large-scale deformations. Our solid harmonic-based rotation-invariant filter framework allows efficient computation and ensures preservation of fine-grained details.
Computer-enhanced microscopy: Deconvolution, High dynamic range fusion, Image stitching, denoising and attenuation correction
Microscopy is limited with respect to dynamic range, field of view and image resolution. Especially in thick tissues signal attenuation and noise additionally degrade image quality and hamper quantitative analysis. We developed tools for image stitching, high dynamic range (HDR) fusion and attenuation correction and also provide basic image deconvolution algorithms.
Rigid and elastic registration using energy minimization and discrete optimization
Many tasks require accurate registration of multiple views of the same specimen or inter-subject registration of different indiviuals. We provide general purpose registration algorithms using continuous energy minimization and discrete optimization based on different similarity measures for bio-medical image registration.
Parametric shapes and generative models for image content description
Abstraction from image data is the key goal of digital image analysis. Therefore description of image content in relation to the anatomy is one of the most important and complicated tasks we try to solve.